Re: Aaron Sloman's "The Irrelevance of Turing Machines to AI" article

From: Wolf Kirchmeir (wwolfkir_at_sympatico.ca)
Date: 08/02/04 

Date: Mon, 02 Aug 2004 12:19:22 -0400

Neil W Rickert wrote:

> Wolf Kirchmeir <wwolfkir@sympatico.ca> writes:
[...]
>>Well, my observations of organisms suggest that this is so. As to "where
>>it comes from": Since in children (for example) the ability to
>>discriminate smller _dissimilarities_ increases with age, it seems
>>reasonable to conclude that the ability "to judge similarity" develops,
>>ie, it doesn't "come from" anywhere.
>
>
> On the one hand you say it is a priori. On the other, you say that
> it changes as the child grows. These seem contradictory. And saying
> that it doesn't come from anywhere is simply evasive.

They don't seem contradictory to me. Why shouldn't some innate (_not_ "a
priori") skill or ability change as the child grows? Some pretty nifty
experiments have shown that a newborn spends more time paying attention
to a face-like pattern than a non-face-like patterm constructed from the
same elements. Then, for a while, the infant smiles at all faces,
familiar or not. Then it "makes strange" - it reacts differently to
familiar and unfamiliar faces. Clearly, the child has developed the
ability to discriminate differences that it did not discriminate before.
How do I know? Because of its changed behaviour.

Now you might want to claim that the child always had the ability to
make these fine distinctions, but there is really no way of proving or
disproving such an assertion. The argument that logic requires such a
prior non-developing ability begs the question.

>> The mechanism is operant
>>conditioning in conjunction with changes within the brain, since it
>>appears that there is a sequence in the development of different
>>discrimination behaviours.
>
> This does not make a lot of sense either. Operant conditioning
> already presupposes some notion of similarity, so cannot be used
> to explain how that notion is acquired.

I don't think operant conditioning pre-supposes some notion of
similarity _in the conditioned organism_. The notion is part of a theory
about what we observe, which means it's an abstraction from behaviour.

>> IOW, as the child becomes capable of new
>>classes of behaviour, its environment shapes those behaviours in
>>specific ways. Eg, language development. Ray Solomonoff's work which
>>I've begun to look at, may provide an abstract model of how this
>>happens. (Thanks, Eray.)
>
>
> You responded to my earlier posts with "similarity". You need to
> either explain where similarity comes from, or withdraw your comments
> on my earlier posts.

I've denied that similarity "comes from" anywhere, IOW, that the phrase
is (at best) misleading.

If the phrase is supposed to imply some locus for similarity, it doesn't
make sense.

If the phrase means some sort of assumption needed to build a coherent
theory, well, similarity comes from the theoriser.

If you are asking how we can start off with no "notion of similarity",
and yet as adults use the notion without qualms, well, it develops. That
is, the child develops its ability to discriminate, and the surrounding
culture teaches it to express its response to discriminating
dis/similarity in language using the terms "like, unlike", etc. (We
actively teach children to use these terms - look at any children's
bookstore for teaching materials. Or recall Sesame Street.)

That there is a "built-in" ability to discriminate dis/similarity is
obvious, for without there would be no behaviour to shape. Is that what
you meant by "comes from", where does the discrimination come from in
the first place? If so, it's built in. It's one of the things babies do
from birth.

IOW, it makes sense to use the phrase "comes fom" as a synonym for
"develop" (one of its meanings, BTW), in which case I've already
sketched an explanation. If you mean something else, well, what do you mean?

[...]
>>Once an organism has responded (however indirectly) to some
>>environemntal situation, it will respond the same way to any situation
>>that is "similar enough."
>
>
> Again, you are presupposing that there exists some magical ability to
> sense similarity, and you are failing to provide a basis for this
> ability. You cannot explain similarity in terms of similarity.

Why should something a new-born baby can do be considered "magical?" I
suppose a new-born horse's ability to walk is also magical, then? Or
new-hatched fish's ability to swim? Or a sunflower's ability to rotate
its flower towards the sun? Or a bacterium's ability to move towards or
away from an acidic area of its environment? These are givens. What's
interesting is how these behaviours can be and are shaped by changes in
the environment.

[...]
>>As for "computational definition of similarity", just look at your spell
>>checker. It "judges similarity" between the incorrectly spelled word and
>>correctly spelled ones.
>
>
> Poor example. Similarity between words or spellings comes from the
> culture. Sure, we form representation of these, and use computation
> over representations. But there isn't some grand "similarity"
> algorithm in use. It is more a matter of coding lots of rules in an
> imperfect attempt to simulate culturally based views of similarity.

You're using "comes from" in a rather vague way here. Do you mean
"supplied by?" Or "developed from?" Or "defined by?" Or what, exactly?
And why should "similarity" be "culturally based?" The only "culturally
based" aspects of similarity I'm aware of relate to which
dis/similarities matter.

Granted, my example is not the best, since the algorithm works on
representations, and I've noticed that people have this funny notion
that representations are somehow different from other inputs to a
program. So someone may well not fully understand the example.

IMO it makes no difference whether the program compares bits of data
stored on a hard drive, or bits of data acquired from a sensor, or
whatever. Either way, the algorithm works on data that exist _outside_
the program. Spell checkers and data comparators use the same basic
idea: compare the bit sequences, and compute a difference number. If
it's "0" (say), the bit sequences are identical (and the spell chekcer
skips the word.) If not, look for bit sequences such that the similarity
number between object and target falls within a certin range, in which
case object and target are "similar" or "similar enough" (and the spell
checker displays these "similar enoggh" words on the screen.) (BTW,
"Smart" appliances also use this kind of logic.)

"Rules derived from culture" don't come into it, actually. They do come
into grammar and style checkers, which is one reason such programs are
such abysmal perfomers compared to spell checkers. I know that Word and
the latest WordPerfect mix style checking with their spell checkers, and
offer a so-called "smart correction" option. I've found it to be too
stupid, and turned it off.

[...]
> So either define [similarity], or stop using it in these discussions. Otherwise
> you are calling upon magic which you admit you cannot implement.

OK, how about this:
"Similarity of A and B" is a number between 0 and 1 computed by
comparing elements of A and B. "0" means (A == B), "1" means (A not==
B), any other number means (A approx== B). (This definition inspired by
Eray's account of "minimal information distance."

This definition presupposes a non-question-begging definition of
"compare." Hah!

Whether this definition has anything to do with what actually happens
within an organism that responds similarly to similar situations is
another question.

Relevant Pages